An Examination of Radar and Lightning Characteristics of ... · the path of the storm. The storm...
Transcript of An Examination of Radar and Lightning Characteristics of ... · the path of the storm. The storm...
P 9.14 24th
Conference on Severe Local Storms
An Examination of Radar and Lightning Characteristics of
the “Atlanta Tornado” of March 14-15, 2008
John M. Trostel*, Jenny L. Matthews, and Colleen Coyle
Severe Storms Research Center, Georgia Tech Research Institute, Atlanta, Georgia
Nicholas W. S. Demetriades
Vaisala, Inc., Tucson, Arizona
1 INTRODUCTION
On the evening of March 14, 2008, an EF2 tornado
swept through the downtown area of Atlanta Georgia.
This event was unusual in that it directly impacted the
downtown area of a major metropolitan area. The
effects of the tornado were immediately broadcast
nationwide as the SEC championship quarterfinals were
being held in the Georgia Dome which lay just south of
the path of the storm. The storm continued through the
heart of downtown Atlanta, causing major damage and
resulting in one fatality. A second, non-tornadic storm
followed the first storm on a slightly more southerly track.
NEXRAD WSR-88D radar data has been examined and
compared with Cloud to Ground (CG) and Cloud to
Cloud (CC) lightning data obtained from the National
Lightning Detection Network (NLDN) for two storms, one
tornadic and one non-tornadic that crossed the Atlanta
metropolitan during this event. A modified version of the
Storm Cell Identification and Tracking (SCIT) algorithm
has been used to correlate specific NLDN flashes with
each cell. Lightning characteristics such as flash rate,
peak current, polarity, and flash type have been
examined for both cells.
The general synoptic situation which generated the
storm cells is described in Section 2. This includes a
short discussion of the surface and upper air maps, as
well as the thermodynamic characteristics of the
atmosphere. The radar characteristics of the storms are
described in Section 3. Data from the National Climate
* Corresponding author address: John M. Trostel, Severe
Storms Research Center, Georgia Tech Research Institute,
Georgia Institute of Technology, Atlanta, GA 30332-0857;
email: [email protected]
Data Center (NCDC) NEXRAD archives has been
examined with both the WDSS-II (Weather Decision
Support System – Integrated Information) and the
GR2Analyst software packages. Lightning stroke and
flash data was obtained for these storms from Vaisala
Inc. from the NLDN. This data set is described in
Section 4. The SCIT algorithm is briefly described in
Section 5, along with the modifications made to the
algorithm to enhance it’s usefulness with these storm
cells. The modified SCIT algorithm allowed the
automated association of NLDN flashes with well
defined individual storm cells. The results of the NLDN
and storm cell association and the differing flash
characteristics of each cell are examined in Section 6.
2 SYNOPTIC SUMMARY
The Atlanta Tornado formed under less than ideal
conditions. The tornado developed from one of two
supercells that formed in northeast Alabama and then
moved into northwest Georgia during the evening hours
of March 14, 2008. The two supercells, including one
which produced an EF2 tornado, formed during the
cooler part of Georgia’s main severe weather season.
Severe thunderstorms that take place during this cooler
part of the season or during the November to October
cool season display rather different characteristics than
storms which form during the main tornado season,
from April through June. These cool season storms are
found to most likely develop during the late evening
(Wasula 2004). These nighttime events are also more
likely to exhibit both low convective available potential
energy (CAPE) values, <1000 J/kg, and exist in high
shear environments with storm relative helicity (SRH)
values greater than 200 m2/s
2 (Guyer and Imy 2006).
Due to the differences in parameters between cool and
warm season tornado environments, the environment
leading to a cool season tornado sometimes can look
less threatening than the warm season tornado
environment (Davies 2006). These cool season
tornadoes can, however, be just as deadly as their
warm season counterparts. Therefore, learning how to
accurately predict these storms and understanding the
associated parameters can be very useful.
The tornado that swept through downtown Atlanta on
the evening of March 14, 2008 exhibited many of these
cool season characteristics and presented a somewhat
subtle synoptic situation. With some of the usual
ingredients needed for severe weather in the warm
season missing, the Atlanta tornado is better explained
using cool season severe weather parameters.
The first clue in determining the circumstances
surrounding the development of this tornado was to
examine the CAPE values as well as the SRH values.
Usually, significantly larger values of CAPE and SRH
than were observed are needed to develop tornadic
severe weather in the warm season (Davies 2006). Cool
season tornadic severe weather, however, can be
instigated by the coupling of significantly lower values of
CAPE with relatively high SRH values.
Figure 1 displays the Peachtree City, Georgia WFO
(KFFC) sounding at 00Z from March 15, 2008.
Peachtree City is located about 30 miles south of the
metropolitan Atlanta area. The 00Z sounding
corresponds to a local time of 8:00pm, about 90 minutes
prior to the touchdown of the tornado at 9:38pm. The
sounding indicated a modest CAPE value of 1030 J/kg.
The 0-3 km SRH values were found to be relatively high
at 215 m2/s
2. Therefore, this situation was characteristic
of a cool season storm development with the low,
modest CAPE values coupled with the high SRH values.
Figure 1: KFFC Sounding for 00Z March 15 2008.
Another informative value to look at is the 0-3 km low-
level CAPE. Moderately high and concentrated CAPE
values over 100 J/kg in the lower levels is a good
indicator of low-level buoyancy and indicates sufficient
warm, moist air for tornado development (Guyer and
Imy 2006). The 0-3 km CAPE value from the KFFC
sounding at 00Z was 106 J/kg. This CAPE value in the
low levels confirmed that this tornado has a good source
of low level moisture. The fairly low LCL level located at
759 m further indicated warm, moist low levels (Guyer
and Imy 2006)
Additional severe weather indices which indicated a
marginally severe environment included the Lifted Index
(LI) at -4.5 and the SWEAT index of 302.8 (Rauber et al.,
2005). A peculiar condition to note is the lack of strong
directional shear in the sounding. The KFFC sounding
displayed westerly winds throughout the vertical with
only changes in wind speeds. These changes were,
however, dramatic, with the speed increasing from 30-
40 kts in the low levels to over 100 kts at upper levels.
The March 14th 12Z (8am local time) and March 15
th
00Z (8pm local time) surface charts as seen in Figure 2
and Figure 3, respectively, set up the synoptic situation
just prior to the tornado touchdown. A large low
pressure system initially located in the Arkansas –
Tennessee area at 12Z moved into the Virginia area by
00Z trailing a cold front.
Figure 2. 12Z March 14 2008: Surface Chart
Careful examination of the 12Z chart in Figure 2 shows
the two supercells starting to form in extreme eastern
Alabama and western Georgia just ahead of the trailing
cold front.
Figure 3. 00Z March 15 2008: Surface Chart
The 500 mb heights and temperature plots for 12Z on
the 14th
and 00Z on the 15th are presented in Figures 4
and 5 below. Cool season tornadoes are known to
exhibit somewhat weak 500mb shortwave troughs as is
seen in Figure 4. The deep 500mb troughs that aid in
the formation of supercells and tornadoes during the
warm season are often absent in the cool season
(Guyer and Imy, 2006).
Figure 4. 12Z March 14 2008: 500 mb Chart
Figures 4 and 5 also show a minor shortwave trough
moving to the east as well as a strong upper level jet
across the gulf states.
Figure 5. 00Z 15 March 2008: 500 mb Chart
The short wave progression from the Louisiana and
Mississippi area into Alabama and Georgia area is even
more evident in the 700 mb plots shown in Figures 6
and 7. Also, strong midlevel winds are apparent in both
charts over the affected area.
Figure 6. 12Z March 14 2008: 700 mb Chart
Mid level moisture, apparent in the 12Z 700 mb chart, in
the Alabama and south Georgia area can be seen to be
absent in the 00Z midlevel chart. This drying is
reflected in the large extent of dry air shown at midlevels
in the KFFC 00Z sounding shown in Figure 1.
Figure 7. 12Z 15 March 2008: 700 mb Chart
The lower level, 850 mb heights, temperatures and
dewpoints are shown for 00Z and 12Z in Figures 8 and
9 below.
Figure 8. 12Z March 14 2008: 850 mb Chart
Figures 8 and 9 depict a moderately strong subtropical
jet with wind speeds of 30 to 35 kts over the area
supplying sufficient moisture ahead of the cold front.
Comparison of the 12Z and 00Z plots shows a
substantial increase in low level moisture ahead of the
cold front in the hours prior to initiation of the supercells.
Figure 9. 00Z 15 March 2008: 850 mb Chart
3 RADAR DATA
NEXRAD radar data, obtained from the Level II archives
at NCDC, was used to track the supercell movement.
Initial observations of the cells movement and
development were made using GR2Analyst software.
A plot of storm relative velocity in Figure 10 shows that
the two main supercells demonstrated a significant level
of low level rotation even as they entered the northwest
part of Georgia. Figure 11 depicts the low level (0.5
degree elevation) reflectivity of the lead tornadic
supercell just prior to touchdown at 0127Z. The cell
shows a pronounced hook and a clear inflow notch.
The corresponding plot of low level storm relative
velocity, Figure 12, shows a fairly well defined rotational
couplet corresponding well to the hook and notch areas.
Figure 10. Storm Relative Velocity at 0102Z
Figure 11. Reflectivity of tornadic cell at 0127Z
Figure 12. Storm relative velocity of tornadic cell at 0127Z
Figure 13. Low level reflectivity and TVS of tornadic cell at 0138Z
Figure 13 shows the low level reflectivity at 0138Z,
about 8 minutes after a tornado warning was issued,
and just about the time of the first reports of the tornado
were received Again, the pronounced inflow region is
easily seen just to the right of the TVS symbol, a green
triangle, plotted by GR2Analyst.
At 9:40pm, the TVS signature continued to moved
eastward through the heart of downtown Atlanta. When
the tornado lifted, it had made a path up to 200 yards
wide and 6 miles long.
4 NLDN LIGHTNING DATA
Both Cloud to Ground (CG) and Cloud to Cloud (CC)
lightning flash and stroke data was acquired from
Vaisala’s National Lightning Detection Network (NLDN).
Data was obtained for the state of Georgia covering the
days of March 14th
and 15th
, 2008.
The NLDN consists of over 100 remote, ground-based
IMPACT sensors broadly distributed over the
continental United States. These sensors
collaboratively detect the location of lightning events
using both direction finding (DF) and time of arrival
(TOA) methods. Typically between 6 and 8 sensors are
used to detect each event.
The data from each IMPACT sensor is collected at the
National Collection Center (NCC) located in Tucson,
Arizona. At the NCC, the data is analyzed to produce
data sets consisting of stroke location, time, polarity,
amplitude and type. Additionally, strokes are grouped
into flashes based on temporal and spatial constraints.
CG flash detection efficiency with the NLDN approaches
95 percent, with a median location accuracy of 500
meters. CC events, typically one or two per flash, are
detected at a lower efficiency, on the order of 10 to 20
percent. In this study, comparisons were typically made
between CG and CC rates between different storms.
This allows the difference in detection efficiencies to
play a lesser role than if the CC and CG rates were
compared within a single storm.
5 SCIT MODIFICATIONS AND DATA
INTEGRATION
The temporal correlation of lightning and tornadic
potential in storm cells has been evaluated using a
slightly modified SCIT algorithm associated with the
appropriate NLDN data. The basic SCIT algorithm was
developed to identify, characterize, and track individual
storm cells in NEXRAD data sets (Johnson, et.al., 1998).
One of the features of the original SCIT algorithm was
its use of seven different reflectivity thresholds. This
study benefited from the use of a smaller set of lower
reflectivity thresholds. The use of lower thresholds
resulted in identified storm cells containing a larger area.
This larger area allowed the inclusion of more lightning
strikes for each identified storm. This, in turn, lead to
better accuracy in evaluating lightning strike frequency,
strength, polarity, and cloud-cloud vs. cloud-ground
lightning strikes over time. It also decreased the
likelihood that an identified storm cell would be
improperly tracked. Although some loss in spatial
resolution results from this modification, the
corresponding benefit is greater temporal accuracy.
The modified SCIT algorithm was run on quality
controlled (QC) reflectivity data output by a neural
network in the WDSS-II system (Lakshmanan, et. al.,
2007). The use of QC reflectivity data is important
because much of the high reflectivity returns that result
from non-precipitating targets such as birds, planes, and
ground clutter has been removed. Without this removal
these non-meteorological targets might be identified as
storm cells. The worst case scenario is that such a non-
meteorological target is contained in a true storm area.
According to the SCIT algorithm, the true storm cell
would then be discarded in order to store the higher
reflectivity “storm cell.” Thus, QC reflectivity not only
prevents false storm cells from being detected, but also
prevents true storm cells from being thrown away.
5.1 Initial NLDN and NEXRAD Correlation using
WDSS-II
The WDSS-II system includes components which allow
the user to ingest both NLDN lightning flash data and
NEXRAD radar data and to present both of these data
sets in a temporally correlated display. An example of
the integration of these two data sets for the March 14-
15 event is shown in Figure 14. Observing the two data
sets correlated in time gave a basic confirmation that
the lightning data appeared to be well correlated to the
two main supercells.
The WDSS-II integration method, while useful for
observing both data sets and their temporal evolution,
proved to be somewhat difficult to use when attempting
to automatically associate lightning flashes with
individual storms. The SCIT numbers produced by
WDSS-II tended to not remain associated with the same
cell as time progressed.
A secondary problem when using the SCIT output from
WDSS-II was that the SCIT itself was represented by a
single point with some associated parameters. Typically
earlier studies have addressed this issue by using an
association radius to correlate lightning flash events with
SCITs. This method seemed somewhat arbitrary and
might sometimes lead to erroneous associations.
Figure 14. WDSS-II & NLDN Correlation
5.2 SCIT Processing
In an effort to address these concerns, as well as to
produce irregularly shaped SCIT areas representative of
the radar reflectivity values, a number of adjustments
were made to the standard SCIT algorithm. These
adjustments were designed to improve its performance
in this specific situation, the tracking of several isolated
supercells in the area of interest.
First, a set of adjustable criteria for identifying 2-D storm
cells (storm cells located in a single elevation slice) was
added to increase the probability that a significant storm
far from the radar would be properly identified. The
increase in vertical distance between sample radar
returns increases with range and thus a storm becomes
less likely to be detected in multiple elevation slices.
The inputs included a required minimum range,
minimum mass, and minimum maximum reflectivity
factor of the 2-D component.
The second adjustment made to the standard SCIT
algorithm involved the discarding of lower reflectivity
thresholds. The conventional SCIT algorithm only
retains the inner most and highest reflectivity
components. Any lower reflectivity component that
contains a higher reflectivity component is discarded
and the highest reflectivity component is stored.
However, the current search method involves searching
for any higher reflectivity component that has its azimuth
extent within that of the lower reflectivity value and its
range extent within its current value. There are cases
where this would result in an improper deletion of a
storm cell. Thus a distance criterion was used to
identify the single component that should be deleted.
The most significant addition to the SCIT algorithm was
the creation of a plan view profile of the shape of each
SCIT identified. The standard SCIT algorithm only
records an estimate of the SCIT region. In the enhanced
SCIT processing, a plan view profile was created by
taking the union of its individual 2-D components.
Furthermore, an option of scaling each 2-D component
about its centroid was made available. This allowed for
small errors in the location of phenomena that occur
near the SCIT perimeter. Once the 2-D components
were scaled, the overlapping components were united
into a single shape profile. In some cases components
did not overlap in the X-Y plane (parallel to Earth’s
surface). In this case the convex hull of the 2
components was taken including any area between the
two components.
5.3 SCIT/NLDN Association
Once the plan profile of each SCIT was created,
lightning flashes were spatially associated with the
appropriate SCIT. Lightning frequency and types were
then plotted versus time. The study was narrowed only
to include SCITs detected for an extended period of
time, more than 20 consecutive scans, in order to
include only the most significant and more developed
storm cells.
A plot showing four detected SCITs, three of which were
detected for greater than 20 scans, and the lightning
flashes associated with each SCIT for a single radar
scan is shown in Figure 15. The black ‘blob’ to the right
is the areal SCIT representing the tornadic supercell,
while the slightly smaller pink ‘blob’ to it’s immediate left
is the areal SCIT representing the second non-tornadic
supercell. The third, red ‘blob’ farther to the left
represents a smaller trailing thunderstorm. The small
yellow ‘+’ signs within each ‘blob’ indicate the reported
location of individual NLDN flashes.
Figure 15. MATLAB SCIT / NLDN Association
6 RESULTS
The MATLAB based SCIT tracking routine followed 3
cells, the tornadic supercell, the non-tornadic supercell,
and a third smaller cell, over more than 20 NEXRAD
scans as they moved across the state of Georgia.
Lightning flash data with locations which corresponded
to each of these three cells was collected and analyzed.
These analyses are presented below. In each of the
plots, the non-tornadic cell is represented by a blue line,
the tornadic cell is represented by a green line, and the
smaller trailing cell is represented by a red line. In each
plot, the radar scan number which most closely
corresponded to the time of reported tornado touchdown
is indicated by a filled triangle on the green, tornadic cell
line.
6.1 SCIT Maximum Reflectivity
In Figure 16, the maximum reflectivity, in dBZ, is plotted
for all three cells as a function of radar scan number.
The two larger supercells, plotted in blue and green,
show similar levels of maximum reflectivity, while the
small cell, plotted in red, initially showed a significantly
lower maximum reflectivity level. No significant change
in maximum reflectivity within the tornadic or non-
tornadic SCIT areas can be seen corresponding to
tornado genesis or touchdown.
Figure 16. SCIT Maximum Reflectivity
6.2 Total Flash Rates
CG lightning has been associated with tornadogenesis,
with peaks in CG rates observed prior to tornado
formation and a relative decrease in conjunction with
touchdown (Perez, et. al, 1997). Similar “lightning
jumps” have been found for “total lightning”, the
combination of CG and CC lightning within a storm
(Goodman, et al. 2005). It is now thought that the
lightning jump prior to the formation of the tornado may
be related to the strong updrafts found during this stage
of storm development (Deierling 2008a, 2008b). Initial
work to characterize the lightning jump phenomenon
has shown some correlation to severe weather events
and may be used with spaceborne detection of total
lightning on the Geostationary Lightning Mapper (GLM)
(Schultz 2008).
Total flash rate, as used in this study, is the combined
rate of CG and CC flashes detected within each SCIT
area during each radar scan interval. A plot of this flash
rate for each of the three cells studied is shown below in
Figure 17. The plot for the large, tornadic supercell
shows a larger overall rate for this storm cell. The
tornadic cell total flash rate also shows a significant
reduction just prior to tornado touchdown. This is
followed by a significant increase shortly after
touchdown.
There seem to be at least one other large dip in total
flash rate for this cell later in its lifecycle unrelated to the
tornado. The second, non-tornadic cell also shows a
smaller, but still distinct, dip in total flash rate at almost
the same time as the dip was observed in the tornadic
cell. The overall rate of total flashes is significantly
smaller for the other two, non-tornadic cells.
Figure 17. Total Flash Characteristics
6.3 CG and CC Flash Rates
Figures 18 and 19 below show CG and CC flashes
detected within each SCIT as a function of radar scan
number. There are interesting differences between the
two figures. The CG rate for the tornadic cell starts at a
relatively high rate and decreases slowly to a more
moderate level. The non-tornadic cell exhibits a similar
moderate level throughout its lifecycle. The smaller cell
shows a much lower CG flash rate than either of the two
supercells.
Figure 18. CG Flashes per Scan
Figure 19, depicting the CC flashes within each SCIT,
exhibits significant differences between the non-tornadic
and the tornadic supercells. The tornadic cell CC flash
rate is much higher than the non-tornadic rate over the
entire lifetime of the storms, although some of this
difference may be due to the difference in areal extent
of the two SCITs. Additionally, the tornadic cell shows a
dramatic decrease in CC rate just prior to and during
tornado touchdown. Both supercells show increases in
CC rates in the middle portions of the storm evolution.
Figure 19. CC Flash Rates
6.4 Positive and Negative Flash Rates
Another aspect of the lightning characteristics examined
was the difference between positive and negative
polarity flash rates between the three cells. In both of
the figures below, the CC and CG flashes are combined
and only the polarity of the flash is used to differentiate
the events.
Figure 20 shows the number negative flashes
associated with each SCIT as a function of radar scan
number. While a possible dip in the number of negative
flashes may have occurred for the tornadic cell, the data
is rather noisy and similar dips can be seen at other
times as well. Generally, the negative flash rate for both
supercells looks similar, while the small cell shows a
much lower rate.
Figure 20. Negative Flash Rates
The positive combined CG and CC flash rate for the
three cells is shown in Figure 21. In this case, the
tornadic cell again shows a much higher rate than the
other two cells. Additionally, the tornadic cell flash rate
decreases immediately prior to touchdown and then
exhibits a dramatic increase after touchdown.
Figure 21. Positive Flash Rates
6.5 Mean Flash Magnitudes
The mean flash magnitude for all type of flashes, both
CC and CG and both negative and positive, occurring
within each SCIT area plotted as a function of radar
scan number is shown in Figure 22. The mean flash
magnitude for the tornadic cell appears, in general, to
be lower than the mean magnitudes for the other two
cells. There may be some indication of a small increase
in magnitude for the tornadic cell just prior to and during
tornado touchdown, followed by a longer period of lower
magnitudes.
Figure 22. Mean Magnitude of Flashes
7 CONCLUSIONS
This event was characterized more by the “cool season”
scenario for tornadic development in the southeastern
United States. The storms developed late in the
evening as relatively low-topped supercells in an
environment with only modest CAPE (just above 1000
J/kg). A critical factor in the development of the storms
was adequate low level moisture and a large amount of
vertical speed shear, with a 0-3 km SRH over 200m2/s
2.
The association of the NLDN flash data with the SCITs
showed interesting differences between the two large
supercells. The tornadic cell had a significant decrease
in both CC and positive flashes just prior to and during
the tornado touchdown. There were some differences
in overall rates between the tornadic and non-tornadic
storms, with the tornadic storm having generally larger
rates of both positive and CC flashes. Some of the
difference in overall rates could be due to some
differences in cell size.
Ongoing work in this area should result in improvements
in the SCIT algorithm and better association with the
NLDN data. Areas being explored include using more
robust methods of special clustering for SCIT
determination and temporal correlation for SCIT tracking.
Additionally, normalization of the NLDN data by SCIT
volume will be included.
8 REFERENCES
Davies, Jonathan M., 2006: RUC Soundings with Cool
Season Tornadoes in “Small” CAPE Settings and the 6
November 2005 Evansville, Indiana Tornado. Extended
Abstracts, 23rd
Conference on Severe Local Storms,
Amer. Meteor. Soc., St. Louis, MO, 4.3.
Guyer, Jared L. and Imy, David A., 2006: Cool Season
Significant (F2-F5) Tornadoes in the Gulf Coast States.
Extended Abstracts, 23rd
Conference on Severe Local
Storms, Amer. Meteor. Soc., St. Louis, MO, 4.2.
Rauber, Robert M., John E. Walsh, and Donna J.
Charleviox., 2005: Severe & Hazardous Weather: An
Introduction to High Impact Meteorology. Second
Edition. Kendall Hunt Publishing Company.
Wasula, Alicia C., Lance F. Bosart, Russell Schneider,
Steven J. Weiss and Robert H. Johns, 2004: Cool
Season Tornadoes in the Southeast United States: A
Climatological and Case Study Perspective. Extended
Abstracts, 20th Conference on Severe Local Storms,
Amer. Meteor. Soc., Seattle, WA, 16.4.
Lakshmanan, V., T. Smith, G. J. Stumpf, and K. Hondl,
2007: The warning decision support system - integrated
information (WDSS-II). Weather and Forecasting, 22,
No. 3, 592-608.
Lakshmanan, V., A. Fritz, T. Smith, K. Hondl, and G. J.
Stumpf, 2007 : An automated technique to quality
control radar reflectivity data. J. Applied
Meteorology. vol. 46 no. 3, pages 288-305
Johnson, J.T., P.L. MacKeen, A. Witt, E.D. Mitchell, G.J.
Stumpf, M.D. Eilts, and K.W. Thomas, 1998: The Storm
Cell Identification and Tracking Algorithm: An Enhanced
WSR-88D Algorithm. Wea. Forecasting, 13, 263–276.
A.H. Perez, L.J. Wicker, and R.E. Orville,
“Characteristics of Cloud-to-Ground Lightning
Associated with Violent Tornadoes,” Weather and
Forecasting, vol. 12, Sep. 1997, pp. 428-437.
Goodman, S. J., R. Blakeslee, H. Christian, W. Koshak,
J. Bailey, J. Hall, E. McCaul, D. Buechler, C. Darden, J.
Burks, T. Bradshaw, and P. Gatlin, 2005: The North
Alabama Lightning Mapping Array: Recent severe storm
observations and future prospects. Atmos. Res., 76,
423-437.
Deierling, W., W. A. Petersen, J. Latham, S. Ellis, and
H. Christian, 2008a: The relationship between lightning
activity and ice fluxes in thunderstorms, Journal of
Geophysical Research, 113, p. D15210.
Deierling, W. and W.A. Petersen, 2008b: Total lightning
activity as an indicator of updraft characteristics, Journal
of Geophysical Research, 113, p. D16210.
Schultz, Christopher J. and Walter A. Peterson, 2008:
The Utility of Lightning Jumps in Severe Thunderstorms
in the Tennessee Valley, Extended Abstracts, Third
Conference on Meteorological Applications of Lightning
Data, Amer. Meteor. Soc., New Orleans, LA, P2.3